A Facial Component based Hybrid Approach to Caricature Generation using Neural Networks

K.H. Lai, E.A. Edirisinghe, and P.W.H. Chung (UK)


Caricature generation, artificial intelligence, neuralnetwork, pattern recognition and drawing style


A caricature is defined as a humorous drawing of a person that exaggerates or distorts certain distinctive features. However the caricatures of the same person created by different artists can be very different, since the drawing styles of artists play an important role [1]. Therefore learning the drawing style of an artist provides the key to the computer based automatic generation of professional caricature. In our previous work [2-3], we proposed an example-based caricature generation system, which proved that neural networks can be used for capturing the drawing styles of caricaturists and successfully generating photo-realistic caricatures. Unfortunately, the quality of resulting caricatures was limited by the enormous demand of computational resources required by neural networks used; hence making the resolution of outputs a trade off for computational power and memory limitations. In this paper, we propose a novel facial component based hybrid approach to resolve the above limitations and further improve the quality of the generated caricatures. Detailed experimental and subjective test results are provided and analyzed. This work is an extension of our previous system [3], which is the first attempt to use neural networks in generating caricatures.

Important Links:

Go Back